OpenAI joins Google, Microsoft and Amazon in creating new 'problems' for Nvidia

OpenAI has unveiled its first custom AI processor, called Jalapeño, in a bid to reduce its dependence on Nvidia's dominance in AI computing hardware. The semiconductor processor, developed in partnership with Broadcom, is designed to deliver superior speed and efficiency for handling inference tasks, which involve executing trained models and processing user queries.

OpenAI aims to deploy Jalapeño across its computing infrastructure by the end of the year, signaling a shift towards engineering proprietary chips alongside competitors like Google, Amazon Web Services (AWS), Microsoft, and Meta.

Jalapeño competes directly with Nvidia's GPUs, which have experienced a surge in market valuation due to their use in both model training and daily inference. However, analysts warn that the rise of hyperscaler-designed silicon poses a growing threat to Nvidia's long-term dominance, especially within the fast-growing inference market.

Independent research firm SemiAnalysis claims that Jalapeño outperforms Nvidia's Blackwell architecture in performance per watt, particularly when considering its incorporation of next-generation HBM4 memory. Despite this, Nvidia's GPUs remain crucial for large-scale model training and frontier AI workloads due to their flexibility and established CUDA software ecosystem.

The trend toward custom chips is evident across the tech industry, with Google expanding its Tensor Processing Units (TPUs), Meta committing to deploying one gigawatt of Broadcom-designed AI processors, and Anthropic investing over $100 billion in AWS infrastructure, including Amazon's proprietary Trainium processors. Additionally, emerging chipmakers like Cerebras, SambaNova, D-Matrix, Etched, and Fractile are advancing specialized AI accelerators.